2006/12/25, Eric Pouech eric.pouech@wanadoo.fr:
Maarten,
nice work on Alsa (which is definitively needed !!)
a couple of comments:
- I know the winmm code (and drivers) is crippled by bad synchronisation
tricks (like cleaning a field in a structure to signal a thread). This is bad (TM). So I'd suggest using here a real synchronisation object.
Well, what I have now works well enough, thread is sure to be killed, which makes life easier. I don't see the point in making code more complicated then it needs to be.
- I wonder if it's a good idea to create another thread for this... but
before merging all existing drivers' threads into a single one, I'd suggest using the fd oriented functions & poll (see snd_pcm_poll_descriptors and friends) instead of snd_pcm_wait, that'll ease up the future work.
I'm not sure if it's a good thing either to create a new thread for it, but in the current driver model there are no other alternatives, you are right it is easy enough to use poll(), but basically all I would be doing is copy the code from snd_pcm_wait(), which does it for me. If anyone wants to use poll() it is easy enough to copy the code 1:1 from the LGPL alsa sound library. However I saw no need for it and to keep the code simple I just used snd_pcm_wait.
See alsa-lib/src/pcm/pcm.c for how snd_pcm_wait is implemented: It just uses poll()
Maarten Lankhorst a écrit :
Instead of using asynchronous callbacks that uses signals, use a seperate thread that can be cancelled, this prevents deadlock issues.
Basically we use snd_pcm_wait() that tells us when enough room is free to commit another buffer, then we commit the previous buffer and make the next buffer ready. .....